Assessing Social Vulnerability to Flood Hazards in the Dutch Province of Zeeland

Social vulnerability refers to the social, economic, demographic, and built characteristics of a community that affect its ability to respond to, cope with, recover from, and adapt to environmental hazards (Cutter et al. 2003; HVRI 2019). In the United States (US), Cutter and colleagues (2003) pioneered methods that assessed social vulnerability at the county level. They selected indicators of economic resources, demographic attributes of residents, and type and density of infrastructure to be aggregated into a single index. Their Social Vulnerability Index (SoVI) applied factor analysis to 42 variables, which clustered along 11 factors accounting for roughly 76% of the variance among over 3000 US counties.

Building on the SoVI approach and methodology, other researchers have included additional variables, applied the analysis to various geographic scales, and used different statistical analysis and aggregation methods (Cutter and Finch 2008; Schmidtlein et al. 2008; Fekete 2010; Finch et al. 2010; Sherrieb et al. 2010; Tapsell et al. 2010; Holand and Lujala 2013; Chen et al. 2013; Tate 2012). Our research team used similar methods to explore the resilience capacity of 52 US coastal counties along the northern Gulf of Mexico (Reams et al. 2012). In that study, we adapted 36 variables from the US SoVI analysis and added seven more to include government spending, environmental factors such as the US Environmental Protection Agency’s Toxic Release Inventory, and mean elevation. The method aimed to improve the statistical analysis offered in Cutter’s SoVI, applying principle axis factoring instead of principal component analysis, for a correlation-based approach rather than a variance-based approach (Reams et al. 2012). While the SoVI method adds factor scores to create essentially an index of an index, selecting the top loading variable of each dimension is a method that offers findings bearing greater actionability than a summed factor score, which may be difficult to interpret. Selecting the top loading variable to represent each dimension may be more logical because principal axis factoring takes a correlation-focused approach that includes all variables before determining their importance rather than the variance-based approach of principal component analysis that uses all variables, seeking both common and unique variance in the data (Reams et al. 2012). Lastly, the additive index aggregation of the SoVI method does not make a priori assumptions about the importance of the factors while eigenvalues offer a statistically determined method to express the importance of each dimension through explained variance (Reams et al. 2012).

Other researchers have examined indicators of economic and social vulnerability among European communities. In Belgium, for example, Coninx and Bachus (2007) combined such indicators with flood risk maps in order to identify communities that are more susceptible. They modified indicators from the Social Flood Vulnerability Index developed by the British Flood Hazard Research Centre (Tapsell et al. 2010). In Norway, researchers compared the results of replication of the SoVI using the same measures developed by Cutter and colleagues in the United States, with measures of social vulnerability selected to be more reflective of conditions in Norway (Holand and Lujala 2013). The two indices calculated for Norwegian municipalities were only moderately correlated, which suggests that contextual factors including the geographic scale of data collected, along with cultural, governmental, and institutional characteristics of a jurisdiction are important considerations in the construction of a social vulnerability index (Holand and Lujala 2013).

In Germany, Fekete (2010) demonstrated an approach to assess and validate social vulnerability for communities along river channels. Fekete’s assessment was similar to the index by Coninx and Bachus (2007) and the Bogardi/Birkmann/Cardona framework (Birkmann 2006), which included indicators of social, economic, and environmental conditions. Fekete applied the analysis to 439 German counties and included 41 indicators of age, gender, ethnic background, education, income, and physical environmental factors. The resulting index was validated through a survey of residents who experienced the 2002 river floods in Germany. Fekete found residents of counties classified as more vulnerable did indeed have more difficulty recovering from the river floods.

We selected the Dutch province of Zeeland to demonstrate the construction of a social vulnerability index for several reasons. First, the province contains an interesting mix of land uses, with some areas being predominately agricultural and rural, and others more densely populated and industrialized. Second, its districts and communities vary in terms of wealth, economic vitality, and population growth. Third, there is variation in environmental conditions relevant to future flood exposure and risks, including mean elevations above sea level. Finally, Zeeland Province was the location of major destruction and loss of life following the great North Sea Flood of 1953. That devastation led to the formation of the modern Dutch flood defense program, including the Delta Works network of flood-control structures.

The province of Zeeland, shown in Fig. 1, is located on the Southwest Delta (SWD) of the Netherlands, where the Rhine, Meuse, and Scheldt Rivers meet the North Sea. The SWD is a tide-dominated delta with relatively low river discharge. This southernmost coastal Dutch province boasts popular holiday destinations, ecotourism, and a rich history of trade, commerce, and agriculture. Zeeland encompasses 1784 km² of land and 51 km² of surface waters and supports a population of roughly 381,400 inhabitants distributed at 214 persons per km² (CBS 2015). Zeeland contains 13 municipalities, divided into 164 municipal districts (wijk) that divide municipalities into parcels of generally similar land use. For example, heavy industry does not exist within dense residential neighborhoods. The districts of Zeeland tend to be geographically larger and less densely populated than the Dutch national average given the many farms and other extractive-industry enterprises.

Zeeland’s land area consists of three delta lobes separated by two estuaries, which provide entry to major European trade arteries. Shipping traffic through the Westerschelde to the Port of Antwerp requires open access through the channel, which is the only Dutch estuary with unmanaged tidal exchange. The Westerschelde exhibits a tidal difference commonly over 4 m (van den Berg et al. 1995). The fragmented province exhibits great exposure to both coastal and river flooding, with almost one-third of the land below sea level and with an average elevation of only 1.1 m (Rijkswaterstaat 2014). Figure 1 illustrates Zeeland’s land areas below sea level (in blue) relative to buildings (in orange) to visualize population centers and areas of high physical vulnerability due to low elevation.

Analyzing the pattern of the data in the rotated matrix aids understanding of how the various indicators of social vulnerability are related to each other. We named each of the seven factors to reflect the common dimension of vulnerability they represent.

The factor analysis of 25 indicators of social vulnerability resulted in seven factors explaining about 66% of the total variance. The factors of social vulnerability in Zeeland are urban density, low income, population change, more females, distance to train stations, fewer self-employed, and more employed in service industries.

The index presented here considers dense commercial, industrial, and residential areas as containing the highest levels of social vulnerability. The districts of the southern lobe of Zeeland earned the highest scores due to the intersection of multiple factors associated with social vulnerability, which include limited access to lifelines (that is, evacuation routes) and greater foreign population. The six most vulnerable districts are located in the municipality of Terneuzen, bordering the Westerschelde, which is the only unfortified estuary of the Netherlands. The map in Fig. 2 shows the spatial distribution of the social vulnerability scores for the districts of Zeeland Province. Index scores range from a low social vulnerability score of 0.20 in Schore of Kapelle, to the highest social vulnerability score, 0.64 in Oudelandse Hoeve of Terneuzen.

Forty-five of the 50 districts with highest social vulnerability scores are located in South Zeeland, with the top six found in Terneuzen, the most populous municipality in Zeeland. Terneuzen is situated with Belgium to the south and the Westerschelde, a major shipping corridor, to the north. Terneuzen’s poorest-scoring district is characterized by high population density, a high proportion of female population, a relatively large number of low-income households, and limited access to lifelines, such as evacuation routes. Terneuzen is home to large-scale chemical industry, and contains several districts reserved strictly for industrial activity, which were omitted from the analysis due to the small resident population of these areas. Terneuzen does not have access to rail, but is connected to the Netherlands by a toll tunnel.

The majority of the districts with lower social vulnerability index scores are located in Zeeland’s central region and the municipality of Tholen. Much of the central lobe of Zeeland is prized for tourism and Middelburg, the historic capital of Zeeland, the central lobe and Tholen are serviced by passenger rail, providing critical lifelines to the rest of the Netherlands in the case of an evacuation.

Two districts in the municipality of Kapelle rank in the top 10 lowest social vulnerability index scores, with the third placing just behind, ranked 30. Kapelle is centrally located in the Zeeland, with access to both the Oosterschelde and the Westerschelde, characterized by low population density, stable population relative to change, and nearby access to lifelines.

When we consider the spatial distribution of these social vulnerability scores across the province, several observations can be noted. For example, a high proportion of the elderly population resides in the beach communities of central Zeeland, where towering dunes contribute some of the highest elevations in the province. As a result, the higher elevation may offer flood protection, and thus reduced physical vulnerability to an otherwise socially vulnerable subgroup of the population. Similarly, the Noord-Beveland districts received index scores indicating greater social vulnerability despite relatively high elevation. One contributing factor is due to its role as a tourist destination, and corresponding greater portion of service industry workers. Noord-Beveland does not have access to rail with road access only to the low-lying delta lobes to the north and south.

It is important to point out that factor analysis groups variables together into different factor dimensions based on the similarity in spatial distributions of the variables; it does not necessarily mean that the factors or the selected variables correspond to vulnerability. We interpret how the variables may contribute to social vulnerability within the specific study area based on the findings of related research about sources of social vulnerability. A more objective approach would require validation of the index either through examining empirical damage data (Cai et al. 2016; Lam et al. 2016) or household survey results from those who have experienced large floods as demonstrated in Fekete (2010). Since the flood damage data were not available in the study area (because the study region has not experienced large-scale flooding since 1953), we were not able to verify the derived index.

Our approach is a simplification of the traditional social vulnerability index calculation (Cutter et al. 2003). Instead of computing the combined factor scores and aggregating them into an index, we selected the top-loading variable to represent a factor, and then computed their relative importance and aggregated them into a single vulnerability index. The method can be modified to add two or more variables from each factor if desired. This simplified approach is meant to make it easier for planners and managers to identify the specific attributes of communities that may make them more vulnerable and to identify how these attributes are related to other contextual conditions.

The inclusion of proximity to public transportation is a measure unique to this social vulnerability analysis. In the case of Zeeland, where vehicle ownership is relatively low compared to bicycle ownership, residents without easy access to trains or major roadways would not be able to evacuate or recover quickly if faced with a large-scale flood. For example, traveling from Terneuzen to the provincial capital city Middelburg via public transportation requires multiple bus transfers and a ferry crossing, resulting in over an hour of travel time each way to move a relatively short distance. The fragmentation in the landscape in Zeeland Province introduces a significant potential for “bottleneck” delays during an emergency evacuation. Local planners, residents, and other stakeholders could consider investments in additional rail lines and/or other public transportation modes to help reduce the social vulnerability of these communities to future large-scale floods.